Most RPC tests fail with DeadlineExceeded - handshake ordering violation

[fasterthanlime]

Summary

58 out of 146 tests are failing with DeadlineExceeded. The root cause appears to be a handshake ordering violation.

Root Cause

From the test output, we can see the actual error:

ERROR rapace_core::session: RpcSession::run: handshake failed e=Status { 
  code: InvalidArgument, 
  message: "[verify handshake.ordering]: first frame must be on channel 0, got 1" 
}

The client is sending a request frame on channel 1 before the handshake completes. The handshake requires the first frame to be a Hello on channel 0.

Affected Tests

All tests that use RpcSession::call() or streaming calls are affected:

• rapace::arc_impl tests
• rapace::result_return_type tests
• rapace::method_ids_generation::test_end_to_end_with_method_ids
• rapace-core::transport_conformance tests (mem_, shm_, stream_*)
• rapace-cell::dispatcher_builder::test_multi_service_dispatch_e2e
• rapace-diagnostics-over-rapace tests
• rapace-http-over-rapace tests
• rapace-http-tunnel integration tests
• rapace-template-engine tests
• rapace-tracing-over-rapace tests

Likely Fix

The issue is probably in how RpcSession is being used - calls are being made before session.run() has completed the handshake. Either:

1. The client API needs to wait for handshake completion before allowing calls
2. The test setup needs to ensure handshake completes before making calls
3. There's a race condition in the session initialization

To Reproduce

cargo nextest run -p rapace arc_impl

Any test that makes RPC calls will fail with the same pattern.